Elastic-consistency optimization method and apparatus under agent hive architecture

Abstract

An elastic-consistency optimization method and apparatus under an agent hive architecture, which method and apparatus relate to the application field of agent hive architecture technology. The method is applied to distributed network communication and an intelligent cloud platform hive architecture, and comprises the following steps: S1, performing network attack modeling, comprising: constructing a false-data injection network attack model xi(k+1)=fi(xi(k)), wherein xi(k) represents information sent by an agent i to a neighboring agent, and fi(xi(k)) represents any update mode; and introducing a consensus mechanism into an agent hive architecture, so as to construct a hierarchical topological structure; S2, agents forming a dominant connected subset assuming that these protected nodes form a network; and S3, performing convergence verification on the elastic-consistency optimization method. Compared with the prior art, the present application can solve the problem of agents under a hive architecture reaching a consensus in an unreliable communication network.

Description

An elastic consistency optimization method and device under an agent honeycomb architecture

[0001] The present application claims priority to the Chinese patent application No. 202411347115.8, filed on September 26, 2024, and entitled "An elastic consistency optimization method and device under an agent honeycomb architecture", the whole content of which is incorporated herein by reference. TECHNICAL FIELD

[0002] The present application relates to the technical field of agent honeycomb architecture, and particularly relates to an elastic consistency optimization method and device under an agent honeycomb architecture. BACKGROUND

[0003] The agent honeycomb architecture mode uses a large number of large model inference agents, which collectively work together to solve a problem, and each agent contributes from its unique perspective. The resulting result shows an embodiment of collective wisdom, which exceeds the achievements that any single artificial intelligence individual can achieve, especially in handling complex data sets and scenarios.

[0004] Although the agent honeycomb architecture mode has advantages in flexibility, adaptability and compatibility of swarm intelligence, it is challenging to solve the problem of how agents under the honeycomb architecture reach consensus in unreliable communication networks. SUMMARY

[0005] The present application is aimed at the deficiencies of the prior art, and provides an elastic consistency optimization method under an agent honeycomb architecture with strong practicability.

[0006] The further technical task of the present application is to provide an elastic consistency optimization device under an agent honeycomb architecture with reasonable design, safety and applicability.

[0007] The technical solution adopted by the present application to solve its technical problems is:

[0008] An elastic consistency optimization method under an agent honeycomb architecture, which is applied to distributed network communication and intelligent cloud platform honeycomb architecture, has the following steps:

[0009] S1, network attack modeling;

[0010] Error data injection network attack model: x i (k+1) = f i (x i (k));

[0011] Wherein, x i (k) represents the information sent by agent i to the neighbor agent, f i(x i (k) represents any update method;

[0012] The intelligent agent honeycomb architecture introduces a consensus mechanism and implements a layered topology.

[0013] S2. Agent Composition: Assuming these protected nodes constitute the dominant connected subset of the network, the resilient consistency optimization method is as follows:

[0014] The input is the set of ordinal numbers of the protected nodes in the network, represented as T = {v i The ordinal set of other nodes is represented as O = {v | i = 1, 2, ..., n1}. i The attack node ordinal set is represented as A = {v | i = n1 + 1, ..., n0} (n0 = n1 + n2). i The state information of the group, |i=n0+1,...,n} and its corresponding state information;

[0015] S3. Convergence verification of the elastic consistency optimization method.

[0016] Furthermore, in step S1, the consensus mechanism involves a virtual leader layer with one or more leaders who do not communicate with each other but only send information to nodes in the first layer. This is achieved by dividing n nodes into m layers, labeled from 1 to m from top to bottom. Nodes in the first layer only send information to the second layer, and nodes in the i-th layer send information to at least one node in the (i+1)-th layer. Information transmission between nodes in the i-th layer can be bidirectional. Assuming each layer has r... i If there are n nodes, then r1 + r2 + ... + r m =n.

[0017] Furthermore, the communication network of the agent cellular architecture uses an undirected graph G = (V, E, A), where each node i of the agent has a scalar state at time k, represented as... The states of all nodes in the system are represented by a vector x = [x1, x2, ..., xn]. n ] T express;

[0018] If a subset G of graph G = (V, E, A) is G d Each one that does not belong to G d A node has at least one neighbor belonging to G. d And belongs to G d All nodes form a connected graph, G d It is a dominant connected subset of G = (V, E, A).

[0019] Furthermore, specifically including:

[0020] S2.1, Each node v iReceive the states of the neighbors to form a set

[0021] S2.2, Node v i By identifying the state information of trusted nodes and combining it with the state information of its own nodes, the minimum threshold is selected. and the maximum threshold

[0022] S2.3, Judgment set S i element x in (k) j Does (k) satisfy...? Constitute set R

[0023] S2.4, each node v i The update law is: x j (k)∈R i (k);

[0024] S2.5, Repeat steps S2.1-S2.4 until |x i (k)-x j (k)|<ε,v j ∈T i .

[0025] Furthermore, in step S3, each protected node or ordinary node v i The minimum threshold is determined based on the information of the protected nodes in the neighborhood and the node's own information. and maximum threshold Set R is obtained through threshold filtering. i (k), node v i In the k-th iteration, according to R i The sources of agents in (k) categorize states into three types: those from a set of trusted nodes. The state x of its own node i (k), and a set from ordinary nodes or attacking nodes.

[0026] Furthermore, according to formula (1), it can be rewritten as:

[0027] in, The information can be represented by a minimum threshold and a maximum threshold, i.e., there exists 0 < ρ. j <1, satisfying the following equation

[0028] Furthermore, formula (2) can be rewritten as follows:

[0029] record then

[0030] where M 11 represents the interaction between trust nodes, M 21 represents the action of trust nodes to ordinary nodes, M 22 represents the action of the self node, the consensus mechanism divides the nodes of the communication network into three layers, which are the first layer composed of trust nodes, the second layer composed of ordinary nodes and the third layer composed of attack nodes.

[0031] Further, for the convenience of analysis, record represents the trust node state and the self state, then

[0032] It is not difficult to obtain that M(k) is a row random matrix. The elements in M(k) are non-negative, and there is a lower bound τ for the non-zero elements. And τ = 1 / (d M +1), d M is the maximum value of the radix of R i (k), that is

[0033] Further, analyze the flexibility consistency of the formula mechanism:

[0034] From the properties of the row random matrix, we have:

[0035] where, is a random vector and is independent of t. In the method based on topology reconstruction, the state of the node will reach a consensus and converge to

[0036] An elasticity consensus optimization device under an agent honeycomb architecture, comprising: at least one memory and at least one processor;

[0037] The at least one memory is used to store a machine readable program;

[0038] The at least one processor is used to call the machine readable program and execute an elasticity consensus optimization method under an agent honeycomb architecture.

[0039] Compared with the prior art, the elasticity consensus optimization method and device under the agent honeycomb architecture of the present application have the following outstanding beneficial effects:

[0040] The agent honeycomb architecture mode of the present application uses a large number of large model inference agents that collectively work together to solve a problem, each agent contributing from its unique perspective. The resulting synthesis demonstrates a manifestation of collective intelligence that surpasses the achievements of any individual artificial intelligence, especially in handling complex data sets and a large number of scenarios. It can solve the problem of reaching consensus among agents in a honeycomb architecture in an unreliable communication network. BRIEF DESCRIPTION OF DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0042] Figure 1 is a schematic diagram of an agent honeycomb structure in a flexible consensus optimization method under an agent honeycomb architecture;

[0043] Figure 2 is a schematic diagram of a consensus mechanism under a honeycomb architecture in a flexible consensus optimization method under an agent honeycomb architecture. DETAILED DESCRIPTION

[0044] In order to make the technical personnel in the art better understand the scheme of the present application, the present application will be further described in detail below in combination with specific implementation manners. Obviously, the described embodiments are only some of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0045] A preferred embodiment is given below:

[0046] As shown in Figures 1-2, the flexible consensus optimization method under an agent honeycomb architecture in the present embodiment has the following steps:

[0047] S1, network attack modeling;

[0048] False data injection (FDI) attacks are easy to implement, have strong concealment and interference characteristics, and pose a potential threat to reaching consensus decisions by multiple agents under a honeycomb architecture. Attackers, under the condition of understanding the current system configuration, tamper with the measurement data of sensors or the control input signals of controllers by careful design, thereby misleading the state estimation of the system and making the state of the system deviate from stability.

[0049] Consider the following false data injection cyber attack model: x i (k+1) = f i (x i (k))

[0050] where x i (k) represents the information sent by agent i to its neighbors, f i (x i (k)) represents an arbitrary update method.

[0051] The consensus mechanism of the agent honeycomb architecture has an important influence. Consider a hierarchical topology structure. As shown in Figure 2, the virtual leader layer has one or more leaders, and there is no communication between the leaders, only information is sent to the nodes of the first layer. By dividing n nodes into m layers, from top to bottom, it is marked as 1~m. The nodes of the first layer only send information to the nodes of the second layer, and the nodes of the i-th layer send information to at least one node of the i+1-th layer. The information transmission between nodes in the i-th layer can be bidirectional. Assuming that there are r i nodes in each layer, then r1+r2+…+r m =n.

[0052] Consider the process of the agent honeycomb architecture suffering from false data injection, by adjusting the original communication topology structure, a consensus mechanism model containing one-way information transmission is constructed. The hierarchical topology structure reduces the influence of attack nodes in the network and improves the topological robustness of the network. The communication network of the agent honeycomb architecture is represented by an undirected graph G=(V,E,A), and the node i of the agent has a scalar state at time k, denoted as The state of all nodes in the system is represented by the vector x=[x1,x2,…,x n ] T .

[0053] If the graph is represented by G=(V,E,A), each node in the subset G d does not belong to G d has at least one neighbor belonging to G d , and all nodes belonging to G d form a connected graph, G d is a dominant connected subset of G=(V,E,A).

[0054] S2, the agent forms a dominant connected subset of the network consisting of protected nodes, and the resilient consensus optimization method is as follows:

[0055] Under the false data injection network attack, the trusted nodes can not be attacked by the attack nodes. The protection measures include: ① Generally improve the firewall security level of part of the nodes, make resource redundancy backup of the nodes, and ensure that the nodes get real and reliable information; ② Use digital signature and data encryption technology between nodes, so that the real information sent by these nodes cannot be tampered with.

[0056] These part nodes constitute the protected nodes which constitute the dominant connected subset of the network, then the consistency optimization method is as follows:

[0057] The input of the method is the ordinal set of the protected nodes in the network represented as T={v i ,|i=1,2,...,n1},the ordinal set of other nodes represented as O={v i ,|i=n1+1,...,n0}(n0=n1+n2),the ordinal set of attack nodes represented as A={v i ,|i=n0+1,...,n} and its corresponding state information.

[0058] Specifically, it includes:

[0059] S2.1, each node v i receives the state of the neighbor, and constitutes a set

[0060] S2.2, the node v i selects the smallest threshold and the largest threshold by identifying the state information of the trusted node combined with the state information of the node itself.

[0061] S2.3, judge whether the element x i (k) in the set S j (k) satisfies constitute a set R

[0062] S2.4, the update law of each node v i is: x j (k)∈R i (k);

[0063] S2.5, repeat steps S2.1-S2.4 until |x i (k)-x j (k)|<ε,v j ∈T i .

[0064] S3, verify the convergence of the elastic consistency optimization method.

[0065] Each protected node or normal node v i , according to the information of the protected nodes in the neighborhood and its own information, determine the minimum threshold and the maximum threshold Threshold screening set R i (k). Node v i In the kth iteration, according to the source of the agent in R i (k), the state is divided into three kinds: the state x of the set of trust nodes, the state x i (k) of the self node, and the state x

[0066] According to formula (1) is rewritten as:

[0067] Where, The information of x j (k) can be expressed by the minimum threshold and the maximum threshold, that is, there exists 0 < p 11 < 1, which satisfies the following equation

[0068] Formula (2) is further rewritten as:

[0069] Let Then

[0070] Where, M 11 represents the interaction between trust nodes, M 21 represents the action of trust nodes to normal nodes, and M 22 represents the action of the self node. The consensus mechanism divides the nodes of the communication network into three layers, which are the first layer composed of trust nodes, the second layer composed of normal nodes and the third layer composed of attack nodes.

[0071] For convenience of analysis, let represent the trust node state and the self state, then

[0072] It is not difficult to obtain that M(k) is a row random matrix. The elements in M(k) are non-negative, and there is a lower bound τ for the non-zero elements. And τ = 1 / (d M +1), d M is the maximum value of the cardinality of R i (k), that is,

[0073] Further, analyze the flexibility consistency of the formula mechanism:

[0074] From the properties of row random matrix, we have:

[0075] wherein, is a random vector, and is independent of t, the state of the above method based on topology reconstruction node will reach a consensus, and converge to

[0076] Based on the above method, the resilient consensus optimization device under the agent honeycomb architecture in the embodiment comprises at least one memory and at least one processor.

[0077] At least one memory for storing machine readable programs;

[0078] At least one processor for calling the machine readable program to execute a resilient consensus optimization method under the agent honeycomb architecture.

[0079] The above specific embodiments are only specific cases of the present application, and the patent protection scope of the present application includes but is not limited to the above specific embodiments. Any technical solution conforming to the above specific embodiments of the present application and any appropriate changes or replacements made by ordinary technical personnel in the art shall fall within the patent protection scope of the present application.

[0080] Although the embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, replacements and variations can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. An elastic consistency optimization method under an agent honeycomb architecture, characterized in that, The method is applied to distributed network communication and intelligent cloud platform honeycomb architecture, and has the following steps: S1, network attack modeling; Error data injection network attack model: x i (k+1) = f i (x i (k)) ; where x i (k) denotes the information sent by agent i to its neighbors, f i (x i (k)) denotes an arbitrary update scheme; The consensus mechanism is introduced into the agent honeycomb architecture, and the hierarchical topology structure is carried out; S2, the agent constitutes a hypothesis that these protected nodes constitute a dominant connected subset of the network, and then the elastic consistency optimization method is as follows: The input is a set of node indices in the network to be protected, denoted as T = {v i , | i = 1, 2,..., n1}, a set of other node indices, denoted as O = {v i , | i = n1+1,..., n0} (n0 = n1 + n2), and a set of attack node indices, denoted as A = {v i , | i = n0+1,..., n}, and their corresponding state information; S3, convergence verification is performed on the elastic consistency optimization method. 2.The method of claim 1, wherein, In step S1, the consensus mechanism has one or more leaders in the virtual leader layer, there is no communication between the leaders, only sends information to the nodes in the first layer, by dividing n nodes into m layers, marked from top to bottom as 1~m, the nodes in the first layer only send information to the second layer, the nodes in the i-th layer send information to at least one node in the i+1-th layer, the information transmission between the nodes in the i-th layer can be bidirectional, assuming that there are r i nodes in each layer, then r1+r2+…+r m =n.

3. The method of claim 2, wherein, The communication network of the agent honeycomb architecture uses an undirected graph G = (V, E, A), and the node i of the agent has a scalar state at time k, denoted as The state of all nodes in the system is denoted by a vector x = [x1, x2, …, xN]T n ] T ; If a subset G of graph G = (V, E, A) is G d Each one that does not belong to G d A node has at least one neighbor belonging to G. d And belongs to G d All nodes form a connected graph, G d It is a dominant connected subset of G = (V, E, A).

4. The method of claim 3, wherein, Specifically, it comprises: S2.1, each node v i Receiving the state of the neighbors, constituting a set S2.2, node v i By identifying the state information of the trusted node, combined with the state information of the node itself, the smallest threshold is selected and a maximum threshold S2.3, judging the set S i the element x in (k) j whether (k) satisfies constitute a set S2.4, for each node v i The update law for v is: x j (k)∈R i (k); S2.5, repeat steps S2.1-S2.4 until |x i (k)-x j (k)| < ε, v j ∈ T i .

5. The method of claim 4, wherein, In step S3, each protected node or ordinary node v i The minimum threshold is determined based on the information of the protected nodes in the neighborhood and the node's own information. and maximum threshold Set R is obtained through threshold filtering. i (k), node v i In the k-th iteration, according to R i The sources of agents in (k) categorize states into three types: those from a set of trusted nodes. The state x of its own node i (k), and a set from ordinary nodes or attacking nodes.

6. The method of claim 5, wherein, Rewriting equation (1) as: wherein The information of the minimum threshold and the maximum threshold can be expressed as 0 < p j <1, satisfying the following equation 7. The method of claim 6, wherein, Equation (2) is further rewritten as: Recall then wherein M 11 represents the interaction between trust nodes, M 21 represents the action of trust nodes to normal nodes, M 22 represents the action of the self node, the consensus mechanism divides the nodes of the communication network into three layers, which are the first layer composed of trust nodes, the second layer composed of normal nodes and the third layer composed of attack nodes.

8. The method of claim 7, wherein, For convenience of analysis record represents the state of the trusted node and the state of itself, then It is not difficult to obtain that M(k) is a row stochastic matrix. The elements in M(k) are non-negative, and there is a lower bound τ for the non-zero elements, and τ = 1 / (d M + 1), d M is the maximum value of the cardinality of R i (k), i.e.

9. The method of claim 7, wherein, Analyzing the elastic consistency of the formula mechanism: From the properties of the row random matrix, we have: wherein is a random vector and is independent of t, the states of the method nodes based on topology reconstruction will reach an agreement and converge to 10. A resilient consistency optimization device under an intelligent agent honeycomb architecture, characterized in that, At least one memory and at least one processor; The at least one memory is used to store a machine readable program; The at least one processor is used to call the machine readable program to execute the method in any one of claims 1 to 9. ​

Images